PI: Van P. Carey
Co-PIs: D. Auslander, J.Y. Chen, M. Frenklach, C. Grigoropoulos, R. Greif, R. Horowitz, P. Marcus, A. Packard, P. Pagni, P. Papadopoulos
The Department of Mechanical Engineering at UC Berkeley has 44 faculty members, 420 undergraduate students and 308 graduate students and is rated consistently among the top programs in the country. In 1995 the National Research Council once-a-decade study ranked Berkeley third among American graduate programs in mechanical engineering. Department faculty are internationally recognized for the high caliber of their research, both in terms of its fundamental importance and its impact on significant applications.
Development of advanced computational models of processes and systems is a central component in numerous department research efforts. In many of these projects, modeling efforts are severely limited by available computing power. The requested Network of Workstations (NOW) of 20 Intel processor machines would tremendously enhance numerous research projects in the Department of Mechanical Engineering. Outlined below are two such important research efforts.
The first proposed project is modeling of hierarchical control architectures for Automated Highway Systems (AHS). A key obstacle in the development of AHS is the need for reliable and robust vehicle control systems. Modeling such controls requires considerable computational power. Some of the current research projects conducted by Professor Horowitz in this area involve modeling AHS with thousands of cars rather than a limited number of vehicles. These problems are ideally suited to a parallel processing computational platform. Other faculty involved in this area are Professors Hedrick and Tomizuka.
A second example project is modeling of microscale thermophysics in materials processing. Professor Frenklach is using molecular dynamic simulations to explore transport and surface chemistry issues associated with chemical vapor deposition of diamond thin films. Professor Carey is employing mesoscale Monte Carlo stochastic simulation schemes to model transport during physical deposition processes. Professor Grigoropouls is preparing to use a stochastic simulation scheme to model vapor transport during laser processing of materials at low ambient pressures. Although simulations of more complicated real processes are currently far beyond the power of a single workstation machine, they are well suited to distributed computing. Fully resolved computations could easily use the entire campus NOW.
The two example research projects described above have the potential for enormous impact. Development of automated highway systems has the potential to revitalize the automotive transportation infrastructure by significantly increasing its capacity and making it both safer and more fuel efficient. As traffic in major metropolitan areas becomes increasingly congested, development of AHS technology is one of the few options that appears likely to reduce current transportation problems. Efforts to model microscale thermophysics during materials processing have the potential to enhance development of new materials and to provide analysis tools that can be used to optimize production processes for advanced materials. Eventually, the molecular simulations used in these studies can be used routinely by process development engineers. This can only be achieved when current research improves the basic models and successfully ports them to a computational platform with the power to fully apply them to real processes.
In these two example projects, the requested Intel NOW will facilitate major advances in model development. In AHS modeling efforts, control models can now be tested on a single workstation with a limited number of vehicles in the model. This provides some insight into the viability of the control scheme, but its scale-up to a bigger system cannot be tested. With the NOW, we will be able to assess control models applied to thousands of cars, which is much more realistic than the smaller number of cars that can be handled by a single workstation machine. Moreover, an AHS is a hierarchical system composed by interconnected hybrid systems, each of which is most effectively modeled by different paradigms (e.g. ODEs, discrete event systems, PDEs, etc.). Only through calculations of this type is it possible to assure that the control scheme can handle the complexities of a very large system. In the molecular simulations used to model microscale thermophysics, single workstations can handle only a limited number of molecules over a restricted number of time steps. Local thermodynamic properties can be inferred from the simulations only if enough particles or molecules can be sampled in each time step to get well-defined statistics. Such complex real systems requires memory and processor speed far beyond those provided by a single workstation. Using the requested Intel NOW, we will be able to extend simulations of this type to more complex systems, making it possible to test models of more realistic systems and explore the features of more complicated processes. This type of research is not possible without access to the distributed computing power provided by the NOW.
Once the NOW is operational, some time will be spent evaluating how algorithms must change to facilitate distributed computing. We expect the distributed computing versions of these models would be operational within six months of the time that the NOW becomes operational.
There are numerous other projects in the Department of Mechanical Engineering that also would benefit in a quantum way from the availability of the requested NOW. These include modeling of continuum transport during processing of novel materials (Prof. Greif); development of optimal control strategies for complex systems with many constraints (Prof. Packard); Contour Dynamical calculations of planetary atmospheres and CFD with pseudo-spectral methods (Prof. Marcus); parallel implementation of finite element codes used in the computer simulation of large-scale solid mechanics problems such as metal forming and vehicular crash-worthiness analysis (Prof. Papadopoulos); modeling of turbulent combustion processes in burners (Prof. Chen); and development of an urban/wildland intermix fire growth model (Prof. Pagni).
February 1999